Risk Analysis

DOI: 10.1111/risa.12317

The Effects of Urban Form on Ambient Air Pollution and Public Health Risk: A Case Study in Raleigh, North Carolina Theodore J. Mansfield,1,∗ Daniel A. Rodriguez,2 Joseph Huegy,3 and Jacqueline MacDonald Gibson1

Since motor vehicles are a major air pollution source, urban designs that decrease private automobile use could improve air quality and decrease air pollution health risks. Yet, the relationships among urban form, air quality, and health are complex and not fully understood. To explore these relationships, we model the effects of three alternative development scenarios on annual average fine particulate matter (PM2.5 ) concentrations in ambient air and associated health risks from PM2.5 exposure in North Carolina’s Raleigh-Durham-Chapel Hill area. We integrate transportation demand, land-use regression, and health risk assessment models to predict air quality and health impacts for three development scenarios: current conditions, compact development, and sprawling development. Compact development slightly decreases (−0.2%) point estimates of regional annual average PM2.5 concentrations, while sprawling development slightly increases (+1%) concentrations. However, point estimates of health impacts are in opposite directions: compact development increases (+39%) and sprawling development decreases (−33%) PM2.5 -attributable mortality. Furthermore, compactness increases local variation in PM2.5 concentrations and increases the severity of local air pollution hotspots. Hence, this research suggests that while compact development may improve air quality from a regional perspective, it may also increase the concentration of PM2.5 in local hotspots and increase population exposure to PM2.5 . Health effects may be magnified if compact neighborhoods and PM2.5 hotspots are spatially co-located. We conclude that compactness alone is an insufficient means of reducing the public health impacts of transportation emissions in automobile-dependent regions. Rather, additional measures are needed to decrease automobile dependence and the health risks of transportation emissions. KEY WORDS: Built environment; public health; transportation

1. INTRODUCTION Globally, ambient air pollution is one of the 10 leading causes of premature mortality and preventable disease.(1) In urban areas, motor vehicle emissions are a leading air pollution source and an important public health risk factor.(2) However, under current practice, transportation and urban planners often do not account for air-quality-related health effects when evaluating alternative transportation infrastructure investments and policies.(3) Recently, public health specialists have promoted formal health impact assessment (HIA) as a process

1 Gillings School of Global Public Health, Department of Environ-

mental Sciences and Engineering, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA. 2 Department of City and Regional Planning, University of North Carolina at Chapel Hill, Chapel Hill, NC, USA. 3 Travel Behavior Modeling Group, Institute for Transportation Research and Education, North Carolina State University, Centennial Campus, Raleigh, NC, USA. ∗ Address correspondence to Theodore Mansfield, Doctoral Student, Gillings School of Global Public Health, Department of Environmental Sciences and Engineering, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA; tel: +919-2593659; [email protected].

1

C 2014 Society for Risk Analysis 0272-4332/14/0100-0001$22.00/1 

2 for encouraging transportation and city planners to consider the health impacts of their decisions.(4) Yet, tools to support HIAs of transportation and urban planning projects are not readily available or easy to use. Hence, most HIAs of transportation projects have employed qualitative assessments.(5) Inadequate understanding of the complex relationships linking urban form, air quality, and public health presents a barrier to improved HIA practice for transportation projects. The relationships among urban form, air quality, and health are complex and not fully understood. Compact and walkable urban forms have been associated with lower per-capita vehicle kilometers traveled (VKT) and thus lower emissions of transportation-related air pollutants.(6–13) While one might expect decreases in VKT and the associated air quality benefits to improve public health, recent research has raised concerns that compact urban forms may increase population exposure to poor air quality and hence elevate incidence rates of airquality-associated illnesses due in part to elevated environmental concentrations of pollutants resulting from more compact urban forms.(14–19) Furthermore, while urban form interventions, such as increased residential density and land-use diversity, have shown promise in decreasing transportation-related air pollutant emissions, some studies have shown that innovations in the vehicle fleet, such as the adoption of hybrid vehicle technology, are more effective policy levers in improving air quality.(18,19) Previous research has demonstrated the efficacy of vehicle fleet innovations in reducing mobile source emissions but has not considered the effects of urban form directly.(20) To better understand the manner in which urban form, air quality, and public health interact, we employ an innovative approach for modeling the relationships between urban form and all-cause mortality associated with chronic exposure to fine particulate matter (PM2.5 ) in ambient air. We examine the effects of three alternative land development scenarios—a compact development scenario, a decentralized development scenario, and an intermediate scenario representing current conditions—on annual average PM2.5 concentrations in air and related health effects in the Raleigh-Durham-Chapel Hill area of North Carolina (NC), a region collectively referred to as “the Research Triangle.” Our modeling approach links a transportation demand model that predicts automobile traffic patterns to

Mansfield et al. a land-use regression (LUR) model that predicts PM2.5 concentrations in air. In turn, the LUR model links to a health risk assessment model (Fig. 1). To our knowledge, this is the first study to link a transportation demand model, an LUR model for PM2.5 concentrations, and a health risk assessment model to explore how urban development patterns might influence traffic patterns, air quality, and public health. The modeling framework we employ may be applicable to urban planners, environmental regulators, and public health practitioners in other regions considering land-use and transportation policies as means to improve health outcomes. This research is timely in providing a further exploration, using a new modeling approach, of the relationships among urban form, air quality, and health in a rapidly growing area of study.

2. METHODS 2.1. Study Region The Research Triangle, consisting of Raleigh, Durham, Cary, Chapel Hill, and a number of smaller municipalities, is a sprawling urban agglomeration in central NC with a 2010 population of 1,589,853 spread over 3,380 square miles.(21) Over the past decade, the Research Triangle was the second-fastest-growing metropolitan region in the United States.(22) Smart Growth America has ranked Raleigh third worst in the United States on an index of urban sprawl and as a result the city earned the dubious nickname “Sprawleigh.”(23) This rapid growth and sprawling development pattern typify urban areas throughout the southeastern United States. The region is highly auto dependent, with nearly 90% of trips occurring in private automobiles.(24) Thirty-eight percent of the study area population lives within one-half of a mile of a freeway or major arterial. Due in part to automobile emissions, the Research Triangle suffers from intermittent poor air quality. In 2010, 117 days were recorded with a moderate Air Quality Index and five days were considered “unhealthy for sensitive groups.”(25)

2.2. Transportation Demand Model Transportation planners in the study area use a regional transportation demand model—the Triangle Regional Model (TRM)—to predict the number,

The Effects of Urban Form on Ambient Air Pollution and Public Health Risk

3

Fig. 1. The modeling approach employed in this research links a transportation demand model (Model 1) to an LUR model of PM2.5 concentrations (Model 2) and a health risk assessment model (Model 3) to investigate the effects of urban form on traffic, air quality, and health. Dashed lines indicate data used only for calibration of the LUR model.

length, and types of trips in the region and the modes (private cars, public transit, and nonmotorized) used for each trip. These predictions are based on structural features of the built environment, including land-use patterns and transportation infrastructure, and household socioeconomic data. The TRM is a macroscopic model (in other words, it simulates the whole traffic system rather than individual vehicles) calibrated using household transportation surveys and validated using work trip distributions estimated by the 2000 Census Transportation Planning Package and the 2006–2008 American Community Survey. The TRM covers 12 counties and divides the study region into 2,678 transportation analysis zones (TAZs)—the geographic units used in transportation demand modeling and constructed by the Census Bureau based on census block information. Of the 2,678 TAZs in the study region, 99 are “point TAZs” used to represent a single employment node, resulting in 2,579 TAZs with a defined spatial extent.(21) The TRM employs four steps typical of traditional transportation demand models: (1) trip generation, (2) trip distribution, (3) mode choice, and (4) trip assignment. The output includes estimates of VKT during different time intervals on 19,575 transportation network links in the study region. The TRM explicitly models traffic on freeways and major roadways (major and minor arterials) between TAZs; however, it does not model traffic on local streets or trips that begin and end within the same TAZ. We use the TRM to estimate VKT during the weekday morning peak

traffic period (6:00 AM–10:00 AM) for the study year (2010). The TRM output consists of a line file containing all links in the transportation system and accompanying traffic counts per link. While the TRM does not provide uncertainty associated with traffic count estimation, we use a coefficient of variation derived in a recent study of traffic demand modeling output sensitivity to uncertainties in input variables to develop distributions for each TRM output variable (see Appendix A).(26) The TRM was executed using TransCAD version 5.0 build 1880. 2.3. Development Scenarios We model the impacts of three regional development scenarios on air quality and public health: current (base case) conditions plus two alternative scenarios, one emphasizing compact development and the other urban sprawl. Rather than projecting growth to the future, we consider two alternative presents, given a hypothetical history of different land-use and transportation policy choices. In effect, this modeling approach enables a controlled experiment to test the effects of alternative land-use patterns alone on air quality and population health. In the sprawl and compact development scenarios, we alter the spatial distribution of population and housing while holding constant aggregate population, household socioeconomic characteristics, and transportation system infrastructure based on 2010 data. Scenario 1: Base Case uses observed land-use, population, and employment data and modeled travel patterns in the Triangle in 2010. While the base

4

Mansfield et al.

Fig. 2. Population spatial distributions (left) and histograms of population density in TAZs (right) for the three scenarios. The compact development scenario moves households from low-density areas to higher-density areas, while the sprawl scenario does the opposite. We enter new TAZ population and employment information into the TRM to estimate traffic for each scenario.

case contains several dense urban cores, the region as a whole is relatively dispersed, with low population densities in large portions of the study area, shown in the top left corner of Fig. 2. This dispersed land-use pattern resulted from historic policy choices in the Triangle, including poor investment in mass transit, pedestrian, and bicycle infrastructure; limited jurisdictional cooperation in enacting land development controls; and strong incentives for economic development in suburban research campuses and industrial parks near the region’s geographic center. Scenario 2: Compact Development emulates how land use in the Triangle might be distributed today if growth management and land conservation policies had been implemented in the past. Examples of such policies include density incentives, transfer of development rights, urban growth boundaries, and targeted investment to concentrate growth in urban cores instead of rural areas. To create this scenario, we first calculate the base case population density in each TAZ in the study region. We then

classify TAZs into quintiles based on population density. Next, we reallocate evenly all households and employment locations in TAZs in the two lowest density quintiles across all TAZs in the highest two quintiles. The middle quintile is unchanged from the base case. Thus, households and workplaces are transferred from noncore areas to urban cores, increasing the population and employment density of core urban areas and leaving many noncore areas uninhabited. Given the low population density across much of the study region, reallocating the lowest density TAZs to existing urban cores increases the average density of already dense areas by only a small amount, even though the percentage of the population living in high-density TAZs increases substantially, as shown in the map in the top center of Fig. 2 and the histogram of population density in the center of the right-most column of Fig. 2. Because the density increase in urban cores is modest, this scenario is conservative and may understate potential reductions in motor vehicle travel that could

The Effects of Urban Form on Ambient Air Pollution and Public Health Risk occur given more aggressive policies to discourage sprawl. Scenario 3: Increased Sprawl amplifies the “hollowing-out” of urban areas typical of post-1950s urban development in the United States. This scenario represents a history of policies supporting increased decentralization, including limited land development controls and rapid expansion of utility service areas. To develop this scenario, we first apply a density cap of 1,121 persons/mi2 (1.75 persons/acre)—the 60th percentile of TAZ household density in the base case. We then relocate all households in TAZs with densities above the cap to TAZs with densities below the cap but having nonzero populations in the base case without moving employment locations to increase the spatial mismatch between housing and employment locations. Let ρ = 1,121 people/mi2 (the density threshold). To reallocate population from TAZs with densities above ρ, we employ the following procedure. First, we divide the TAZs into three groups: (1) those with density >ρ, (2) those with density ࣘρ, and (3) those with zero population. Let Mi denote the population in TAZi from within the first group of TAZs. Let Rj represent the population that TAZj (from within the second group of TAZs) could receive without exceeding a density of ρ. Then for each TAZi in group 1, we remove Mi persons, and into each TAZj in group 2, we add an additional number of persons equal to:  Mi (1) Rj × i j Rj Fig. 2 maps the resulting population spatial distributions (left) and histograms of TAZ population densities (right) for the three scenarios. In the sprawl scenario, a relatively high percentage of the population resides in TAZs with relatively low population densities whereas in the compact growth scenario, a much higher percentage of the population lives in TAZs with relatively high population density. Comparing the population density histograms between scenarios (right), a significant portion of the base case population is shifted to higher density TAZs in the compact scenario and to lower density TAZs in the sprawl scenario. 2.4. LUR Model To quantify the effects of land-use changes on PM2.5 concentrations, we develop and calibrate an

5

LUR model for the study region. Previous studies have found that LUR models perform well in estimating PM2.5 and other air pollutant concentrations at fine spatial scales, with R2 values of 0.6–0.7.(27–32) At fine spatial scales, LUR models have significant advantages over more complex chemical transport models due to their computational simplicity.(33–35) Additionally, LUR models can represent pollutant spatial heterogeneity that may not be accounted for by geostatistical techniques (such as kriging) in cases where monitoring station locations mask spatial heterogeneity due to their intentional siting away from major local pollution sources.(36) A meta-analysis of studies assessing the spatial extent of pollutants from mobile sources found that significant spatial gradients exist in PM2.5 concentrations above regional background levels in urban areas.(37) Thus, LUR is an appropriate and demonstrated technique to capture expected spatial variations in urban PM2.5 concentrations above regional background concentrations. We rely on previous LUR studies to identify explanatory variables most likely to significantly predict PM2.5 concentrations in urban areas. While the study region contains a limited number (eight) of fixed-site air quality monitors, studies investigating the region-to-region transferability of LUR models suggest that observations from as few as 10 monitoring stations are sufficient to calibrate an LUR model that has previously been successfully calibrated in an urban area with similar cultural and regulatory environments.(27,38–40) Furthermore, research suggests that the variation between stations is more important for model calibration than the number of stations.(41) While our study region contains only one rural monitor, the outcome variable and all explanatory variables tested, with the exception of industrial land use, vary significantly among monitors (Table II). 2.5. Calibration Data Fig. 3 shows the locations of the eight air quality monitoring stations in the study area. Seven stations track both PM2.5 and PM10 mass concentration over time, with most stations tracking daily concentrations since 1999. The eighth station tracks only PM10 concentrations and has records dating back to 1990. Consistent with previously demonstrated methods, annual average PM2.5 concentrations are estimated at the latter monitoring station using the average PM2.5 :PM10 ratio observed at the other seven monitoring stations.(42) Four monitoring stations stopped

6

Mansfield et al.

Fig. 3. Air pollution monitoring station locations.

reporting PM2.5 concentrations before 2010, but time series data from prior years are available (Table I). For these stations, we developed station-specific linear regression models including all available time series data in order to predict 2010 PM2.5 concentrations. Factors that influence PM2.5 concentrations, including urban form, transportation demand, and vehicle fleet characteristics, vary over time; thus, we must impute missing 2010 data to capture changes in these factors over time. Table I shows the resulting regression models and their significance levels. Three of the four models (monitors 183–0015, 135–0007, and 063–0001) are statistically significant at the 95% confidence level. The fourth model (183–0003) is significant only at the 85% confidence level; however, we retain this station due to limited data in the study region. Additionally, cross-validation (Section 3.1) showed that inclusion of station 183–003 in the analysis did not significantly change the parameters of our LUR model, supporting the decision to include these data in our analysis. Common explanatory variables in studies predicting PM2.5 concentrations include traffic density and/or intensity, land-use classification, population density, and elevation.(28–32) We fit a linear LUR model to the study area by testing combinations of the following explanatory variables: weekday AM peak VKT, population density, and total industrial land use as defined in the most recent available land-use data.(43–49) The value of each explanatory variable is calculated within four circular buffers (0.5, 1, 1.5, and 2 km) around each monitoring station.

Elevation differences between monitoring stations used to calibrate the LUR model are minimal, as the last column of Table II shows; therefore we do not include elevation as an explanatory variable. While variation in elevation around a station may increase vehicle emissions (due to increased emission rates when vehicles travel uphill), we did not identify previous LUR studies in the literature that found variation in elevation around air quality monitoring stations a significant predictor of annual average PM2.5 concentrations. Table II summarizes all explanatory data within all buffer sizes used for model calibration. 2.6. Risk Assessment Model As the first step in assessing the health risks of PM2.5 air pollution under each development scenario, we impose a 1 km × 1 km grid containing 9,185 cells over the study area. As Fig. 1 illustrates, the LUR model predicts annual average PM2.5 concentration, denoted as Cik for grid cell k within TAZ i. Then, for each TAZ i, we compute the spatial average PM2.5 concentration, denoted as Ci , as the spatially weighted mean value of the Cik s. Like numerous previous studies quantifying impacts of PM2.5 on health,(42,50–54) the following health impact function translates Ci into an estimate ࢞yi of the number of premature deaths in 2010 in TAZ i attributable to PM2.5 : yi = yio × AFi = yio ×

(2)

2010 2010 2010

183–0020

037–0004 063–0015

12 4

3

4 11 10 9 12

n(Years of Data)

−0.396**

29.1**

n/a

F1,10 = 24.33* −0.728 F1,2 = 6.8 PM2.5 :PM10 ratio, all stations:

n/a

17.47**

F1,2 = 54.7* F1,9 = 3.1 F1,8 = 21.9** F1,7 = 23.4** F1,10 = 70.9**

−0.567* −0.181 −0.244** −0.320** −0.520**

20.26* 16.98** 16.59** 18.30** 20.50** n/a

F-Statistic

β (μg/m3 Yrs)

α (μg/m3 )

16.8 88.0 48.3 8.89 47.2 9.26 77.5 36.0 41.5 28.0

77.7 145 109 14.0 96.4 18.8 145 104 88.7 47.0

1.5 164 220 175 36.6 146 26.0 207 166 143 68.0

2 5,080 2,850 4,280 309 3,860 147 4,360 2,120 2,880 1,750

0.5

Note: Bold values indicate explanatory variable values used in the final LUR model.

1.91 16.2 8.91 3.98 8.66 0.10 21.0 13.6 10.4 6.64

183–0015 183–0003 183–0014 183–0020 135–0007 037–0004 063–0001 063–0015 Mean SD

1

4,480 2,650 3,670 309 4,900 161 4,080 2,360 2,830 1,700

1

4,060 2,540 3,120 310 5,180 170 3,880 2,380 2,710 1,650

1.5

2 3,480 2,530 2,750 749 4,400 173 4,150 2,530 2,600 1,400

Buffer Size (km)

Buffer Size (km)

0.5

Populating Density (Persons/mi2)

Weekday AM Peak VKT (1,000s of VKT)

Air Quality Station ID

0.77 0.64

0.95 0.18 0.70 0.74 0.86

Adj. R2

Table II. Summary of All Explanatory Variables Tested

**p < 0 .01; *p < 0 .05;  p < 0.15. n/a = too few observations. a Station does not track PM . 10 b Station does not track PM . 2.5

2003 2000 2008 2007 2010

Most Recent Year

183–0015 183–0003 135–0007 063–0001 183–0014

Air Quality Station ID

8.9 10.2

10.0

n/a n/a n/a n/a 10.2

Observed

0 0 0 0 0 0 4 6 1.3 2.2

0.5

0 0 15 0 10 0 17 82 16 26

1

0 0 36 30 13 0 34 86 25 27

1.5

Buffer Size (km)

20 0 63 38 15 6 42 86 34 28

2

127 125 100 120 145 121 147 118 125 14

Elevation (m)

n/aa 0.634 0.673

n/aa

n/aa n/ab n/aa 0.684 0.668

Average PM2.5 : PM10 Ratio

Industrial Land Use (Acres)

9.5 10.1

n/a

8.9 13.4 11.7 11.9 10.2

Predicted

2010 Value (μg/m3 )

Model Form: PM2.5 = α + β × Years After 1990

Table I. Regression Models and PM2.5 :PM10 Ratios for Estimating PM2.5 Concentrations in Cases Where Data Are Missing

The Effects of Urban Form on Ambient Air Pollution and Public Health Risk 7

8

Mansfield et al. ∞ Ci=o

∞ RR(Ci )P(Ci )dCi − Ci=o RR(Ci )P (Ci )dCi ∞ , Ci=o RR(Ci )P(Ci )dCi

where ࢞yi is deaths per year attributable to PM2.5 in ambient air in TAZ i, yi0 is total deaths in TAZ i in the year 2010, AFi is fraction of deaths attributable to PM2.5 in ambient air in TAZ i, Ci is annual average PM2.5 concentration in TAZ i (in μg/m3 ), RR(Ci ) is relative risk of premature mortality when exposed to PM2.5 at an annual average concentration Ci , equal to 1 + 0.06Ci /(10 μg/m3 ) (from Ref. 56); P(Ci ) is a probability distribution of population exposure to PM2.5 in TAZ i under the scenario being modeled, and P’(Ci ) is the counterfactual exposure distribution— assumed to represent the case in which all VKT = 0. Total deaths from all-cause mortality in 2010 are calculated for each TAZ using 2010 population and county-level death rates.(55) Because the relative risk function is assumed to be linear and the exposure concentration in each TAZ is assumed to be constant across the population, Equation (2) simplifies to the following:

yi = yio × AFi = yio ×

(Ci −α) 10 0.06 × C10i

0.06 × 1+

,

(3)

where α is the LUR model constant, assumed to represent regional background PM2.5 concentrations. This health risk assessment approach, although the standard approach used in quantifying health impacts of PM2.5 in ambient air, makes several important simplifying assumptions. First, the approach uses household location as a proxy for exposure. In the epidemiological studies that underlie the health impact function in Equation (2), exposures at other locations, such as workplaces and schools, are averaged across large populations.(56) We assume exposures outside the household are similar in our application and thus use household exposure as a proxy for total exposure as well, which is common practice in other risk assessments of ambient air.(42,50–54) Finally, we assume that 2010 death rates are applicable across all scenarios; the relationship between land use and transportation behavior remains stable across all scenarios; population density is uniform within TAZs; and PM2.5 exposure concentrations are uniform within TAZs.

3. RESULTS 3.1. LUR Model We calibrate the LUR model using the observed 2010 annual average PM2.5 concentrations (where available) or predicted concentrations (where observed data are not available) shown in Table I. We test all unique combinations of buffer sizes and explanatory variables to maximize the adjusted R2 of the model. Neither acres of industrial land use nor household density significantly predicted PM2.5 concentration, regardless of buffer size; thus, they are not included in the final model. The final LUR model (adjusted R2 = 0.80; root mean square error = 0.61 μg/m3 ; F1,6 = 28.4, p = 0.0018) for annual average PM2.5 concentrations in the study area is: PM2.5 = 8.69 + 0.0473 × V KT,

(4)

where PM2.5 is predicted annual average PM2.5 concentration (μg/m3 ) and VKT is thousands of VKT traveled during the AM peak within a 1,000 meter buffer of the center of the estimation cell. Both the model constant (p < 0.0001; standard error = 0.445 μg/m3 ) and the VKT parameter (p = 0.0011; standard error = 8.89×10−3 μg/m3 per thousand VKT) are statistically significant. Due to data limitations previously discussed, the model was calibrated using a small sample (n = 8). Given the small sample size, we employ a standard leave-one-out cross-validation (LOOCV) procedure for all monitoring stations to test the robustness of the model. The LOOCV shows that the model is quite robust despite limited availability of observational data (Table III), with a low root mean square error and high correlation between predicted and observed concentrations. Extrapolation beyond the range of observed independent variable values can be a concern for LUR models. However, in this study, VKT values outside the observed range (9.26–88.0) occurred for only 0.6%, 0.7%, and 0.3% of the study area in the base case, compact, and sprawl scenarios, respectively. The model is limited due to poor variation in industrial land use around monitoring stations (Table II). While the model is unable to capture variation in PM2.5 concentrations attributable to industrial or special (e.g., airport) land uses, this limitation is not necessarily relevant given the fundamental research aim of investigating how transportation behavior,

The Effects of Urban Form on Ambient Air Pollution and Public Health Risk

9

Table III. Results of LOOCV for Each Monitoring Station Air Quality Station ID (Station Removed)

Model Constant (μg/m3 )

VKT Parameter (μg/m3 VKT)

8.90 8.89 8.76 8.25 8.62 8.79 8.61 8.72

0.0445 0.0392 0.0483 0.0541 0.0466 0.0458 0.0516 0.0472

37–183–0015 37–183–0003 37–183–0014 37–183–0020 37–135–0007 37–037–0004 37–063–0001 37–063–0015

a Time-series b PM

2.5 :PM10

Predicted PM2.5 (μg/m3 )

Actual (Imputed) PM2.5 (μg/m3 )

9.7 12.3 11.1 8.7 10.8 9.2 12.6 10.4 Root MSE (μg/m3 ): 0.84 Correlation of predicted to observed: 0.82 (p < 0.05)

Square Error (μg2 /m6 )

8.9a 13.4a,b 10.2 10.0 11.7a 8.9 11.9a 10.2

0.5 1.2 0.8 1.7 0.7 0.1 0.6 0.1

imputed value. imputed value. Table IV. Area and Population Exposed to PM2.5 Concentrations Above EPA and WHO Standards EPA Standard (12 μg/m3 )

WHO Standard (10 μg/m3 )

Area Exceeding Standard, mi2 (%)

Population Exposed (%)

Area Exceeding Standard, mi2 (%)

Population Exposed (%)

24.4 (0.7%) 33.9 (1.0%) 21.9 (0.7%)

35,500 (2.2%) 95,400 (6.0%) 14,100 (0.9%)

236 (7.0%) 249 (7.4%) 256 (7.6%)

407,000 (25.6%) 644,000 (40.5%) 191,000 (12.0%)

Base case Compact development Sprawl

Table V. VKT, Population Density, and Percent of Total Population by VKT Decile Mean VKT (Thousands)

Mean Population Density (Persons/mi2 )

Percent of Total Population (%)

VKT Decile

Base

Compact

Sprawl

Base

Compact

Sprawl

Base

Compact

Sprawl

1 2 3 4 5 6 7 8 9 10

0.0 0.2 0.5 1.0 1.5 2.3 3.8 6.7 14.1 42.8

0.0 0.0 0.1 0.3 0.7 1.3 2.6 5.9 13.6 44.4

0.0 0 1.2 2.2 3.3 4.9 7.0 10.4 17.3 44.5

121 118 138 170 197 271 376 496 1,057 1,687

48 40 40 52 87 138 307 471 1,060 2,232

276 302 332 359 376 419 460 540 640 714

2.6% 2.6% 3.0% 3.7% 4.2% 5.9% 8.1% 10.7% 22.8% 36.4%

1.1% 0.9% 0.9% 1.2% 1.9% 3.1% 6.9% 10.5% 23.7% 49.9%

6.2% 6.8% 7.5% 8.1% 8.5% 9.5% 10.4% 12.2% 14.5% 16.2%

air quality, and health impacts respond to changes in regional land-use patterns.

3.2. PM2.5 Concentrations In each grid cell, we use the TRM’s traffic volume estimates for each roadway link to tabulate

total VKT during 6:00–10:00 AM occurring within a 1 km buffer of the centroid of the cell and apply the LUR model to estimate annual average PM2.5 concentrations (Fig. 3). As expected based on the LUR model, the highest estimated annual average PM2.5 concentrations are generally located near significant links in the transportation network: in the base case, the highest estimated concentration,

10 15.0 (9.61–41.3) μg/m3 , occurs at the intersection of two limited-access highways that provide regional mobility. While the 1 km buffer surrounding this grid cell comprises only 0.036% of the total study area, 0.62% of the predicted VKT occurs within the same buffer, illustrating how local impacts may arise due to regional transportation patterns. Estimated PM2.5 concentrations in the compact development scenario are more spatially concentrated; however, the region-wide average estimated concentration is nearly identical to the base case (Table VI). Thus, while impacts are more localized, region-wide concentrations are relatively unchanged despite lower predicted transportation demand. The highest estimated concentration of PM2.5 occurs in the same cell as in the base case; however, in the compact development scenario, the estimated concentration in this cell increases slightly to 15.2 (9.57–48.8) from 15.0 μg/m3 . This result is intriguing in that potentially conflicting policy goals at the local and regional levels arise: while compact development may reduce VKT—and in turn may reduce vehicle emissions—air quality issues may become pronounced in specific locations. Relative to the base case, the sprawl scenario reduces localized impacts; however, the regionwide average PM2.5 concentration increases slightly (Table VI). The highest estimated concentration of PM2.5 occurs in the same cell as in the base case and the compact development scenario; however, the maximum estimated concentration in the increased sprawl scenario decreases to 14.3 (9.30–39.4) from 15.0 in the base case and 15.2 μg/m3 in the compact development scenario. Thus, while dispersing population across the study area reduces the highest predicted PM2.5 concentrations, regional average PM2.5 concentrations are slightly higher. While not estimated directly, vehicle emissions are an important source of PM2.5 ; therefore, in conjunction with predicted increases in VKT, we can infer that the sprawl scenario increases aggregate vehicle emissions relative to the base case and compact development scenarios. Several limitations of the LUR model should be noted. We calibrated the model using 2010 data and did not consider meteorological data; thus, the model implicitly assumes 2010 meteorological conditions and can therefore only be used to estimate PM2.5 concentrations in 2010. However, this limitation does not affect comparisons between scenarios. Additionally, we assume that the spatial distribution of the modeled weekday morning peak traffic

Mansfield et al. patterns sufficiently represents the spatial variation in traffic throughout the year. Despite these limitations, diagnostics of the LUR model suggest that it captures a large amount of observed variation in PM2.5 concentrations given 2010 meteorological conditions. In all three scenarios, portions of the study area exceed both the EPA standard for annual average PM2.5 concentrations (12 μg/m3 ) and the World Health Organization (WHO) air quality guidelines for annual average PM2.5 concentrations (10 μg/m3 ).(57,58) The fine-grained spatial scale of our air quality predictions enables comparisons across scenarios of the number of people living in areas that exceed both the EPA and WHO standards for annual average PM2.5 concentrations (Table IV). Relative to the base case scenario, compact development increases the number of people living in areas above the EPA and WHO thresholds by 59,962 and 236,913 persons, respectively, whereas the sprawl scenario decreases these exposures by 21,346 and 279,238 persons, respectively. Similarly, the compact development scenario increases the total area exceeding the EPA and WHO standards by 9.5 and 13.6 mi2 , respectively. In contrast, the sprawl scenario decreases the total land area exceeding the EPA standard by 2.5 mi2 but increases the area exceeding the WHO standard by 20.3 mi2 (6.7 mi2 more than in the compact development scenario). Thus, while more compact development may result in higher numbers of people living in more polluted areas, more dispersed urban forms may decrease the spatial extent of relatively higher PM2.5 (i.e., greater than 12 μg/m3 ) concentrations and increase the spatial extent of moderate PM2.5 concentrations (i.e., 10 to 12 μg/m3 ). 3.3. Attributable Mortality Fig. 4 presents estimated death rates per 100,000 persons attributable to exposure to PM2.5 above regional background for all scenarios. For the base case, the model estimates 47 (5–167) deaths in 2010 associated with PM2.5 exposure (Table VI). In the compact development scenario, estimated premature mortality associated with PM2.5 exposure increases to 65 (6–220). The compact development scenario increases PM2.5 exposure despite a marginal reduction in regional PM2.5 concentrations (Tables IVand VI); however, the point estimate of associated mortality is within the 95% confidence interval (CI) for the base case. Compared to the base case, population density is relatively higher in areas with higher predicted

The Effects of Urban Form on Ambient Air Pollution and Public Health Risk

11

Table VI. Vehicle Miles Traveled, Annual Average PM2.5 Concentrations, and Attributable Mortality by Development Scenario (Mean and 95% CI) PM2.5 Concentration

Attributable Mortality

VKT Total (Thousands) Mean (μg/m3 ) Maximum (μg/m3 ) Total (Persons) Max. Rate (per 100k Persons) Scenario 1: Base Case Scenario 2: Compact Development Relative to Base Case Scenario 3: Increased Sprawl Relative to Base Case

21,300 (2,600–80,200) 20,200 (2,700–76,800) −5.2% 26,700 (2,600–96,700) +25.2%

9.0 (7.8–10.7) 9.0 (7.8–10.7) −0.2% 9.1 (7.9–10.9) +1.0%

VKT in the compact development scenario despite aggregate reductions in transportation demand (Table VI). In the sprawl scenario, 31 (2–122) deaths are associated with PM2.5 exposure—less than in the base case, but once again within the bounds of the 95% CI. It is counterintuitive that sprawling development reduces the point estimate of mortality associated with exposure to PM2.5 considering higher predicted regional VKT and higher predicted PM2.5 concentrations over much of the study area compared to the base case scenarios (Table VI; Fig. 4). However, dispersing population reduces household exposure enough to counteract the modest increase in the regional PM2.5 concentrations. Table V shows that those living in areas with high traffic (TAZs in the highest decile of VKT; last row of the table) decrease substantially in the sprawl scenario compared to the base case. In contrast, in the compact development scenario, half of the population lives in areas with very high traffic. Across scenarios, the areas with the highest predicted death rate associated with PM2.5 generally share two characteristics: (1) close proximity to major regional transportation corridors and (2) high population density. Thus, while clustering population near transportation infrastructure may reduce transportation demand, our study indicates that this effect may not be powerful enough to counteract the negative health effects associated with concomitant increased exposure to PM2.5 . Furthermore, Fig. 4 illustrates that regional mobility needs may result in highly localized PM2.5 hotspots near major regional transportation corridors; if development is clustered near these hotspots, highly localized health impacts may also result. While this finding is intuitive, the tendency of highly localized health impacts to persist

15.0 (9.6–41.3) 15.2 (9.6–48.8) +1.7% 14.3 (9.3–39.4) −4.4%

47 (5–160) 65 (6–220) +39.1% 31 (2–120) −33.1%

17 (2.1–85) 20 (2.8–90) +21.0% 17 (1.7–68) +0.82%

despite significant land-use change underscores the need to address transportation issues holistically when considering health outcomes. Several limitations should be considered in interpreting these results. Because we consider transportation infrastructure, including public transit, fixed across all scenarios, it is likely that the compact development scenario overpredicts VKT. Compact development would likely be complemented by increased provision of local and express bus services and increased walkability, resulting in shifts to public transit and nonmotorized transport (cycling and walking) that may not be captured by the TRM. However, this simplification likely results in minimal error given the conservative nature of the compact development scenario: while the scenario increases population density in urban cores, the magnitude of the increase is not necessarily large enough to trigger significant investment in public transit. Furthermore, VKT in the sprawl scenario is likely underestimated. Many rural areas have few transportation links modeled by the TRM; thus, only a small portion of the redistributed rural population may live near a transportation network link with predicted use. Furthermore, the TRM does not predict intrazonal trips and trips on minor roads; thus, shortdistance motorized and nonmotorized trips are likely systematically underpredicted. While the low population density in the sprawl scenario likely results in a low number of short nonmotorized trips, it also likely leads to a larger number of short motorized trips that are not captured by the TRM, resulting in unaccounted-for VKT from intrazonal trips that may be larger in the sprawl scenario than in the other two scenarios. This effect is magnified in TAZs with large geographic extents, which are more heavily

12

Mansfield et al.

Fig. 4. Annual average PM2.5 concentrations and attributable mortality for each scenario.

populated in the sprawl scenario. Additionally, this research does not consider many potential health benefits of compact development, such as increased physical activity from increased walking and cycling.

While there is emerging interest in characterizing how increased physical activity and increased exposure to air pollutants interact to influence health outcomes in urban areas, we do not consider these

The Effects of Urban Form on Ambient Air Pollution and Public Health Risk potential interactions. Finally, while the efficacy of vehicle fleet characteristics in reducing transportation emissions and improving air quality has been demonstrated,(17–19) we do not consider changes in vehicle fleet characteristics. As the adoption of improved vehicle technology in the fleet increases over time, the negative health impacts associated compact development relative to other urban development patterns will likely be reduced. Overall, while this study supports recent findings that compactness may exacerbate health impacts,(17,18) these results should be interpreted in the context of studies that find positive air quality impacts from other interventions that may require population density above certain thresholds, such as increased investment in public transportation, and positive health impacts from increased physical activity that may be associated with walkable and transit-supportive urban forms.(59) 3.4. Sensitivity Analysis To consider the effects of uncertainty in key model variables on the results of this research, we perform a sensitivity analysis, focusing on four sources of uncertainty: (1) the LUR model constant (representing the regional background PM2.5 concentration); (2) the LUR model parameter (which predicts how VKT changes affect PM2.5 concentration); (3) estimated VKT; and (4) the relative risk of premature mortality for 10 μg/m3 increases in PM2.5 concentration. We repeat our analysis while varying one of these model inputs at a time, using the upper and lower bounds of the 95% CI for each variable: 8.24–9.13 μg/m3 for the LUR model constant; 0.0385–0.0562 μg/m3 per 1,000 VKT for the LUR model parameter; the upper and lower bounds of VKT in each TAZ accounting for uncertainty and variability (see Appendix A); and 1.02–1.11 for the relative risk per 10 μg/m3 in PM2.5 concentration. Fig. 5 shows that the health risk estimates are most sensitive to uncertainty in the relative risk parameter. In addition, the estimates are sensitive to assumptions about VKT in each TAZ. Nonetheless, the differences among scenarios—the fundamental research question being explored—are stable at both the low and high ends of the 95% CIs for these variables. By contrast, the LUR model constant and VKT parameter in the LUR model have comparatively small effects on the risk estimates. In sum, the sensitivity analysis shows that: (1) the relative risk function is the greatest source of uncertainty in

13

estimating mortality attributable to PM2.5 and (2) the differences among scenarios are generally consistent when random model input variables range across their 95% CIs, although the magnitude of these differences changes in some instances. This latter finding strengthens the conclusion that compact development may exacerbate health impacts if pursued without other strategies to reduce automobile dependency. 4. DISCUSSION Compact development alone may not be an effective strategy for reducing the health impacts of exposure to PM2.5 in urban areas, even though it may reduce regional VKT. If households are clustered near significant transportation corridors to achieve compact urban forms, an increasing proportion of the population may be exposed to high concentrations of PM2.5 unless complementary incentives to reduce trip length, encourage use of public and nonmotorized transit, and/or increase the adoption of lower emitting vehicle technologies are provided. Thus, other policies should be considered, alone or in conjunction with compact development, to reduce the health impacts of transportation—a conclusion consistent with studies finding that road pricing strategies may reduce exposure to pollutants in ambient air while land development controls may increase exposure and that an urban sprawl land development scenario may decrease exposure to pollutants for individuals who move to the urban periphery but increase exposure for individuals who remain in urbanized areas, compared to a base case scenario.(9,18) Our findings also support recent literature suggesting that neighborhood-scale air quality is an important risk factor for a variety of negative health outcomes and that holistic policy approaches are critical in improving health in urban areas.(60) While not considered here, if compactness is associated with spatial homogeneity of income or races, there is potential for disproportionate impacts on specific populations, raising equity and environmental justice concerns.(61) While our findings support decentralization as a means of reducing air-quality-related public health impacts in a limited sense, this result must be strongly qualified. Given existing urban forms, the relatively slow nature of land-use change, and existing large-scale transportation infrastructure that supports private automobile mobility, simply avoiding development in areas with locally high PM2.5

14

Mansfield et al.

Fig. 5. Relative effect of varying each model component from its lower to upper 95% CI value on estimated mortality for each scenario.

concentrations due to established regional mobility patterns is not a feasible policy option. Furthermore, decentralized development may also decrease opportunities for physical activity, increase emissions of pollutants that exacerbate global climate change, and diminish ecosystem services with indirect human health benefits via increased land consumption—all outcomes that likely have negative impacts on human health at both local and global scales. In a broader sense, an intriguing finding is the potential for counterintuitive outcomes when considering human health. The complex relationships that link urban form, transportation behavior, air quality, and public health merit continued research to better inform policymakers with the goal of improving public health in urban areas. Considering the rising use of HIA in the United States, our findings support the integration of quantitative methods into HIA practice to untangle complex relationships and elucidate tradeoffs that occur in real-world decision-making environments. A second thought-provoking finding is the critical role that large transportation infrastructure investments may play in influencing the health impacts of transportation in urban areas. Despite significant land-use change across scenarios, we find the location of and degree to which PM2.5 hotspots are

above regional average concentrations remarkably consistent. The consistency of PM2.5 hotspots despite significant changes in regional land use is likely attributable to the hierarchal nature of transportation systems, which tend to have limited redundancy and funnel large proportions of regional traffic through specific network links. Thus, existing transportation infrastructure may play a significant role in characterizing the relationships linking land use, transportation behavior, air quality, and public health—and ultimately constrain the ability of certain policy instruments to positively affect public health outcomes. The potential of transportation investments to influence how land use, transportation behavior, and air quality interact with health outcomes supports the integration of quantitative HIA into long-range scenario-planning efforts to test the sensitivity of health outcomes to future land-use visions in the context of future transportation infrastructure investments. ACKNOWLEDGMENTS This research was supported in part by funds from the NIH (T32ES007018). We would like to personally acknowledge Dr. Paul Voss for providing

The Effects of Urban Form on Ambient Air Pollution and Public Health Risk

15

Table A1. Summary of TAZs by Quintile of Population Density Base Case

Population density quintile

Compact Development

Sprawl

Mean VKT

CV

Mean VKT

CV

Mean VKT

CV

1

37,940

0.53

51,760

0.47

26,002

0.68

2 3 4 5

30,310 22,767 14,592 12,431 53,496 25,439

0.70 0.99 1.38 1.60 0.50 0.95

35,944 31,723 22,886 20,280 19,164 27,237

0.56 0.68 0.99 1.29 1.45 0.99

33,328 27,763 17,612 14,918 50,528 25,545

0.61 0.80 1.12 1.28 0.50 0.87

Zero pop. TAZs All TAZs Note: CV = Coefficient of variation.

valuable statistical consultation and Monisha Khurana for assisting with ArcGIS analysis supporting this research. We would also like to acknowledge four anonymous reviewers who provided thoughtful comments that improved the quality of this article.

APPENDIX A: CHARACTERIZATION OF UNCERTAINTY AND VARIABILITY IN VKT ESTIMATES To consider potential sources of uncertainty and variability in the VKT estimates that underlie our LUR model, we consider: (1) VKT uncertainty given uncertain trip generation within the TRM and (2) variability in VKT within groups of similar TAZs. To account for uncertainty within the TRM, we assume that the uncertainty in the TRM output is characterized by a normal distribution having a mean equal to the TRM output and an SD equal to the mean multiplied by the coefficient of variation (CV) derived by Zhang et al. in a study characterizing the uncertainty in model outputs from a four-step transportation demand model using Monte Carlo simulation to account for uncertain model inputs (Fig. A1).(26) To account for variability, we first remove all TAZs with zero population and then divide the remaining TAZs into quintiles by population density for each scenario. We then calculate the CV for each quintile and the zero-population group for each scenario (Table A1). We assume that the variability in VKT within each TAZ is characterized by a lognormal distribution with a mean equal to the mean predicted by the TRM (adjusted for uncertainty) and an SD equal to the mean multiplied by the CV calculated for each group of TAZs.

Fig. A1. Analytica model schematic.

We combine the VKT distributions described above with other model parameters, assuming distributions as appropriate, in Analytica (Lumina Decision Systems, Los Gatos, CA, USA) to estimate health risks (Fig. A1; Table A2). All distributions are truncated as appropriate to avoid spurious values in the tails of the distribution (e.g., VKT is truncated to avoid values less than zero). We use the VKT distributions obtained from above, combining both uncertainty and variability, to estimate PM2.5 in each TAZ using our calibrated LUR

16

Mansfield et al. Table A2. Analytica Model Variables

Variable Name

Variable Type

Specification

TAZ Mean VKT (VKT) CV from uncertainty (CVU ) SD from uncertainty (SDU ) VKT with uncertainty (VKTU )

Index Discrete; indexed by TAZ Discrete Expression Normal; indexed by TAZ

CV from variability (CVV ) SD from variability (SDV ) VKT (VKTU+V )

Discrete Expression Lognormal; indexed by TAZ

Beta0 (β 0 )

Normal

Beta1 (β 1 )

Normal

PM PM above Relative risk function (RRF)

Expression; indexed by TAZ Expression; indexed by TAZ Normal

Relative risk (RR) Attributable fraction (AF) Baseline death rate (DR; per 100,000 persons) Population (pop)

Expression; indexed by TAZ Expression; indexed by TAZ Discrete; indexed by TAZ

Mortality (mort) PM death rate (DRPM ; per 100,000 persons) Sum of mortality

Expression; indexed by TAZ Expression; indexed by TAZ

Range: 1–2,579 TRM outputs (data not shown) 0.4245(26) CVU × VKT Mean: VKT SD: SDU See Table A1 CVV × VKT Mean: VKTU SD: CVV Mean: 8.689 SD: 0.555 Mean: 0.0474 SD: 0.011 β 0 + β 1 × VKTU+V PM – β 0 Mean: 1.06(56) SD: 0.023 1 + ((PM˙above/10)*(RRF − 1)) (RR − 1)/(RR) 2010 death rates by county(55) (data not shown) TRM input data or scenario specifications (data not shown) AF*Pop*DR/100,000 Mort/pop*100,000

Expression

Sum(mort, TAZ)

Discrete; indexed by TAZ

model (Equation (4)). We assume both the model constant and the VKT parameter are normally distributed. We then subtract background PM2.5 concentrations from these estimates and apply Equation (3), assuming that the RR function is normally distributed. REFERENCES 1. Lim S, Vos T, Flaxman D, Danaei G, Shibuya K, AdairRohani H, Amann M, Anderson HR, Andrews KG, Aryee, M, Atkinson C, Bacchus LJ, Bahalim AN, Balakrishnan K, Balmes J, Barker-Collo S, Baxter A, Bell ML, Blore JD, Blyth F, Bonner C, Borges G, Bourne, R, Boussinesq M, Brauer M, Brooks P, Bruce NG, Brunekreef B, Bryan-Hancock C, Bucello C, Buchbinder R, Bul Fl, Burnett RT, Byers TE, Calabria B, Carapetis J, Carnahan E, Chafe Z, Charlson F, Chen H, Chen JS, Cheng AT-A, Child JC, Cohen A, Colson KL, Cowie BC, Darby S, Darling S, Davis A, Degenhardt L, Dentener F, Des Jarlais DC, Devries K, Dherani M, Ding EL, Dorsey ER, Driscoll T, Edmond K, Eltahir Ali S, Engell RE, Erwin PJ, Fahimi S, Falder G, Farzadfar F, Ferrari A, Finucane MF, Flaxman S, Francis Gerry R Fowkes Greg Freedman, Freeman MK, Gakidou E, Ghosh S, Giovannucci E, Gmel G, Graham K, Grainger R, Grant B, Gunnell D, Gutierrez HR, Hall W, Hoek HW, Hogan A, Hosgood HD, Hoy D, Hu H, Hubbell BJ, Hutchings SJ, Ibeanusi SE, Jacklyn

GL, Jasrasaria R, Jonas JB, Kan H, Kanis JA, Kassebaum N, Kawakami N, Khang Y-H, Khatibzadeh S, Khoo J-P, Kok C, Laden F, Lalloo R, Lan Q, Lathlean T, Leasher JL, Leigh J, Li Y, Lin JK, Lipshultz SE, London S, Lozano R, Lu Y, Mak J, Malekzadeh R, Mallinger L, Marcenes W, March L, Marks R, Martin R, McGale P, McGrath J, Mehta S, Memish ZA, Mensah GA, Merriman TR, Micha R, Michaud C, Mishra V, Hanafiah KM, Mokdad AA, Morawska L, Mozaffarian D, Murphy T, Naghavi M, Neal B, Nelson PK, Nolla JM, Norman R, Olives C, Omer SB, Orchard J, Osborne R, Ostro B, Page A, Pandey KD, Parry CDH, Passmore E, Patra J, Pearce N, Pelizzari PM, Petzold M, Phillips MR , Pope D, Pope CA, Powles J, Rao M, Razavi H, Rehfuess EA, Rehm JT, Ritz B, Rivara FP, Roberts T, Robinson C, RodriguezPortales JA, Romieu I, Room R, Rosenfeld LC, Roy A, Rushton L, Salomon JA, Sampson U, Sanchez-Riera L, Sanman E, Sapkota A, Seedat S, Shi P, Shield K, Shivakoti R, Singh GM, Sleet DA, Smith E, Smith KR, Stapelberg NJC, Steenland K, ¨ Stockl H, Stovner LJ, Straif K, Straney L, Thurston GD, Tran JH, Dingenen RV, Donkelaar Av, Veerman JL, Vijayakumar L, Weintraub R, Weissman MM, White RA, Whiteford H, Wiersma ST, Wilkinson JD, Williams HC, Williams W, Wilson N, Woolf AD, Yip P, Zielinski Jan M, Lopez AD, Murray CJL, Ezzati M A comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor clusters in 21 regions, 1990–2010: A systematic analysis for the Global Burden of Disease Study 2010. Lancet, 2013; 380:2224–2260. 2. Mayer H. Air pollution in cities. Atmospheric Environment, 1999; 33(24–25):4029–4037.

The Effects of Urban Form on Ambient Air Pollution and Public Health Risk 3. Wernham A. Health impact assessments are needed in decision making about environmental and land-use policy. Health Affairs, 2011; 30(5):947–956. 4. National Research Council. Improving Health in the United States: The Role of Health Impact Assessment. Washington, DC: National Academy Press, 2011. 5. Bhatia R, Seto E. Quantitative estimation in health impact assessment: Opportunities and challenges. Environmental Impact Assessment Review, 2011; 31(3):301–309. 6. Frank L, Sallis J, Conway L, Chapman E, Saelens E, Bachman W. Many pathways from land use to health—Associations between neighborhood walkability and active transportation, body mass index, and air quality. Journal of the American Planning Association, 2006; 72(1):75–87. 7. Song J, Webb A, Parmenter B, Allen D, McDonald-Buller E. The impacts of urbanization on emissions and air quality: Comparison of four visions of Austin, Texas. Environmental Science and Technology, 2008; 42(19):7294–7300. 8. De Ridder K, Lefebre F, Adriaensen S, Arnold U, Beckroege W, Bronner C, Damsgaard O, Dostal I, Dufek J, Hirsch J, IntPanis L, Kotek Z, Ramadier T, Thierry A, Vermoote S, Wania A, Weber C. Simulating the impact of urban sprawl on air quality and population exposure in the German Ruhr area. Part I: Reproducing the base state. Atmospheric Environment, 2008; 42:7059–7069. 9. De Ridder K, Lefebre F, Adriaensen S, Arnold U, Beckroege W, Bronner C, Damsgaard O, Dostal I, Dufek J, Hirsch J, IntPanis L, Kotek Z, Ramadier T, Thierry A, Vermoote S, Wania A, Weber C. Simulating the impact of urban sprawl on air quality and population exposure in the German Ruhr area. Part II: Development and evaluation of an urban growth scenario. Atmospheric Environment, 2008; 42:7070–7077. 10. Marshall J. Energy-efficient urban form. Environmental Science and Technology, 2008; 42(9):3133–3137. 11. Behan K, Maoh H, Kanaroglou P. Smart growth strategies, transportation and urban sprawl: Simulated futures for Hamilton, Ontario. Canadian Geographer, 2008; 52(3):291–308. 12. Transportation Research Board (TRB) of the National Academies of Science. Driving and the Built Environment: The Effects of Compact Development on Motorized Travel, Energy Use, and CO2 Emissions – Special Report 298. Washington, DC: Transportation Research Board, 2009. 13. Koracin K, Bassett S, Mouat D, Gertler A. Application of a scenario-based modeling system to evaluate the air quality impacts of future growth. Atmospheric Environment, 2009; 43:1021–1028. 14. Clark L, Millet D, Marshall J. Air quality and urban form in U.S. urban areas: Evidence from regulatory monitors. Environmental Science and Technology, 2011; 45:7028–7035. 15. Marshall J, Brauer M, Frank L. Healthy neighborhoods: Walkability and air pollution. Environmental Health Perspectives, 2009; 117(11):1752–1759. 16. Schweitzer L, Zhou J. Neighborhood air quality, respiratory health, and vulnerable populations in compact and sprawled regions. Journal of the American Planning Association, 2010; 76:363–371. 17. Hixson M, Mahmod A, Hu J, Bai S, Niemeier D, Handy S, Gao S, Lund J, Sullivan D, Kleeman M. Influence of regional development policies and clean technology adoption on future air pollution exposure. Atmospheric Environment, 2010; 44:552–562. 18. McDonald-Buller E, Webb A, Kockelman K, Zhou B. Effects of transportation and land use policies on air quality. Transportation Research Record, 2010; 2158:28–35. 19. Stone B, Mednick A, Holloway T, Spak S. Mobile source CO2 mitigation through smart growth development and vehicle fleet hybridization. Environmental Science and Technology, 2009; 43(6):1704–1710. 20. Frey HC, Zhai H, Rouphail, NM. Regional on-road vehicle running emissions modeling and evaluation for conventional

21.

22. 23. 24. 25. 26.

27.

28.

29.

30.

31.

32.

33.

34.

35.

36. 37. 38.

39.

17

and alternative vehicle technologies. Environmental Science and Technology, 2009; 43(21):8449–8455. Triangle Regional Model Service Bureau. Triangle Regional Model Version 5: Model Documentation Report, 2010. Available at: http://www.campo-nc.us/TRM/Version-5Model-Documentation.pdf, Accessed October 31, 2013. Kotkin J, Cox W. The census’ fastest-growing cities of the decade. Forbes Magazine, 2011. Goldberg S. Reigning in Sprawleigh. Time Magazine, 2011. United States Census Bureau. 2010-Year American Community Survey Estimates, 2010. Available at: http://www. factfinder2.census.gov/, Accessed October 11, 2012. United States Environmental Protection Agency. AirData Website, 2012. Available at: http://www.epa.gov/airdata/, Accessed May 5, 2012. Zhang T, Xie C, Waller S. Integrated equilibrium travel demand model with nested logit structure: Fixed-point formulation and stochastic analysis. Transportation Research Record, 2011; 2254:79–96. Briggs D, Hoogh C, Gulliver J, Wills J, Elliot P, Kingham S, Smallbone K. A regression-based method for mapping trafficrelated air pollution: Application and testing in four contrasting urban environments. Science of the Total Environment, 2000; 253:151–167. Hoek G, Beelen R, Hoogh K, Vienneau D, Gulliver J, Fischer P, Briggs D. A review of land use regression models to assess spatial variation of outdoor air pollution. Atmospheric Environment, 2008; 42:7561–7578. Hoek G, Beelen R, Kos G, Dijkema M, Van der Zee S, Fischer P, Brunekreef B. Land use regression model for ultrafine particles in Amsterdam. Environmental Science and Technology, 2011; 45(2):622–628. Henderson S, Beckerman B, Jerrett M, Brauer M. Application of land use regression to estimate long-term concentrations of traffic-related nitrogen oxides and fine particulate matter. Environmental Science and Technology, 2007; 41(7):2422– 2428. Ross Z, Jerrett M, Ito K, Tempalski B, Thurston G. A land use regression for predicting fine particulate matter concentrations in the New York City region. Atmospheric Environment, 2007; 41:2255–2269. Moore D, Jerrett M, Mack W, Kunzli N. A land use regression model for predicting ambient fine particulate matter across Los Angeles, CA. Journal of Environmental Monitoring, 2007; 9:246–252. Potoglou D, Kanaroglou P. Carbon monoxide emissions from passenger vehicles: Predictive mapping with an application to Hamilton, Canada. Transportation Research Part D, 2005; 10:97–109. Zwack L, Hanna S, Spengler J, Levy J. Using advanced dispersion models and mobile monitoring to characterize spatial patterns of ultrafine particles in an urban area. Atmospheric Environment, 2011; 45:4822–4829. Greco S, Wilson A, Hanna S, Levy J. Factors influencing mobile source particulate matter emissions-to-exposure relationships in the Boston urban area. Environmental Science and Technology, 2007; 41(22):7675–7682. Levy, J, Hanna S. Spatial and temporal variability in urban fine particulate matter concentrations. Environmental Pollution, 2011; 159(8–9):2009–2015. Zhou, Y, Levy J. Factors influencing the spatial extent of mobile source air pollution impacts: A meta-analysis. BMC Public Health, 2007, 7:89 Vienneau D, de Hoogh K, Beelen R, Fischer P, Hoek G, Briggs D. Comparison of land use regression models between Great Britain and the Netherlands. Atmospheric Environment, 2010; 44:688–696. Jerrett M, Arain A, Kanarouglou P, Beckerman B, Potoglou D, Sahsuvaroglu T, Morrison J, Giovis C. A review and evaluation of interurban air pollution exposure models. Journal

18

40.

41. 42.

43. 44.

45.

46. 47. 48.

49.

50.

Mansfield et al. of Exposure Analysis and Environmental Epidemiology, 2005; 15(2):185–204. Poplawski K, Gould T, Setton E, Allen R, Su J, Larson T, Henderson S, Brauer M, Hystad P, Lightowlers C, Keller P, Cohren M, Silva C, Buzzelli M. Intercity transferability of land use regression models for estimating concentrations of nitrogen dioxide. Journal of Exposure Science and Environmental Epidemiology, 2009; 19:107–117. Ryan P, LeMasters G. A review of land use regression models for characterizing intraurban air pollution exposure. Inhalation Toxicology, 2007; 19:127–133. Cohen A, Anderson H, Ostro B, Pandey K, Krzyzanowski M, Kunzli N, Smith K. The global burden of disease due to outdoor air pollution. Journal of Toxicology and Environmental Health Part A. 2005; 68(13–14):1301–1307. Chatham County Geographic Information Systems Mapping. Chatham County Zoning Map, 2011. Available at: http://www.ftp.chathamnc.org/Zoning/, Accessed May 5, 2012. Durham County Geographic Information Systems Department. Durham County Parcel Data, 2012. Available at: http://www.durhamnc.gov/ich/as/ts/Pages/DistributionPolicies.aspx, Accessed May 5, 2012. Pittsboro Planning Department. Town of Pittsboro Zoning Map, 2011. Available at: http://pittsboronc.gov/vertical/sites/ %7B512CE168--4684--4855--9CD9--7D209FE775E3%7D/ uploads/Pittsboro ZONING MARCH 2013.pdf, Accessed May 5, 2012. Orange County Geographic Information Systems Division. Orange County Parcels, 2010. Available at: http://www.web. co.orange.nc.us/gisdownloads, Accessed May 5, 2012. Town of Carrboro Geographic Information System. Carrboro Zoning, 2012. Available at: http://www.gis.ci. carrboro.nc.us/Carrboro/Zoning/, Accessed May 5, 2012. Town of Chapel Hill Geographic Information System Data and Services. Chapel Hill Zoning Districts, 2008. Available at: http://www.gis.townofchapelhill.org/download data/, Accessed May 5, 2012. Wake County Geographic Information System Mapping Services. Wake County Property Data, 2012. Available at: http://www.wakegov.com/gis/Pages/maps.aspx, Accessed May 5, 2012. Fann N, Lamson A, Anenberg S, Wesson K, Risley D, Hubbell B. Estimating the national public health burden associated with exposure to ambient PM2.5 and ozone. Risk Analysis, 2012; 32:81–95.

51. Levy JI, Greco SL, Melly SJ, Mukhi N. Evaluating efficiencyequality tradeoffs for mobile source control strategies in an urban area. Risk Analysis, 2009; 29(1):34–47. 52. MacDonald Gibson J, Brammer A, Davidson C, Folley T, Launay F, Thomsen J. Environmental Burden of Disease Assessment: A Case Study in the United Arab Emirates. Dordrecht, The Netherlands: Springer, 2013. 53. MacDonald Gibson J, Thomsen J, Launay F, Harder E, DeFelice N. Burden of disease attributable to environmental pollution in the United Arab Emirates. PLOS ONE, 2013; 8(3):e57536. 54. Li Y, MacDonald Gibson J, Jat P, Puggioni G, Hasan M, West J, Vizuete W, Sexton K, Serre M. Burden of disease attributed to anthropogenic air pollution in the United Arab Emirates: Estimates based on observed air quality data. Science of the Total Environment, 2010; 408(23):5784– 5793. 55. North Carolina State Center for Health Statistics. Detailed Mortality Statistics, North Carolina Residents, 2010. Available at: http://www.schs.state.nc.us/schs/deaths/dms/2010/, Accessed October 31, 2013. 56. Pope C, Burnett R, Thun M, Calle E, Krewski D, Ito K, Thurston G. Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. Journal of the American Medical Association, 2002; 287(9):1132– 1141. 57. National Ambient Air Quality. Final Rule, 78 Federal Register 3086–3287, January 15, 2013. 58. World Health Organization. WHO Air Quality Guidelines for Particulate Matter, Ozone, Nitrogen Dioxide and Sulfur Dioxide: Summary of Risk Assessment. Geneva: World Health Organization, 2006. 59. Bauman E, Reis R, Sallis J, Wells J, Loos R, Martin B. Correlates of physical activity: Why are some people physically active and others not? Lancet, 2012; 380:258– 271. 60. Hankey S, Marshall J, Brauer M. Health impacts of the built environment: Within-urban variability in physical inactivity, air pollution, and ischemic heart disease mortality. Environmental Health Perspectives, 2012; 120(2):247– 253. 61. Rodriguez D. The Influence of Compact Development on Travel Behavior and Tailpipe Emissions, a Case Study of Mecklenburg County in 2050. Invited Presentation, Transportation Energy and Environment Meeting, 2010.